Updated on 15 November 2026, by João Leite, Founder of murmur.

A GEO course, to appear in AI answers, teaches four things: how AI models search, the bot, the cited page and measurement. GEO, short for Generative Engine Optimization, is the work of putting the company inside the answers of ChatGPT and the other AI models, and each of the four parts has official, free material. The first is the simplest to check at the source: OpenAI writes, in the help center article about search in ChatGPT, that “ChatGPT search typically rewrites your query into one or more targeted queries that it sends those providers.” Anyone who understands that sentence understands why the right page is the one that answers the question the AI asks, and not just the word the person typed.

  • How AI models search. ChatGPT rewrites the question into one or more searches and sends them to partner search engines, according to OpenAI.
  • The bot that reads the site. OpenAI asks that the site not block OAI-SearchBot so that the content enters ChatGPT’s summaries and snippets.
  • The page that becomes a source. In the study by Aggarwal and colleagues, citing sources, adding quotations and adding numbers increased the page’s space in the AI answer by 30% to 40%, in one of the study’s measures.
  • Measurement. Otterly saw the variation in measured presence drop by about 73% when going from 10 to 100 questions, in a set of 320.
  • The course teaches, and the work takes time and data. Applying the four parts at the same time, every week, in the company’s market, is continuous work.

A GEO course teaches four things: how AI models search, the bot, the page that becomes a source and measurement

A well-built GEO course covers four parts, and each one answers a practical question for anyone who wants the company in AI answers. How AI models search answers where the answer draws companies from. The bot answers whether the AI can read the site. The page that becomes a source answers what to write. Measurement answers whether the work is producing results.

The four parts depend on one another. The best-written page does not enter the answer if the bot cannot read it, and a readable site does not appear if no page answers the buyer’s question. Without measurement, nobody knows which of the parts failed.

A course that talks about only one of them, such as “writing for AI” or “configuring robots.txt”, teaches one part of the work. Before paying, it is worth checking the syllabus to see whether all four appear, and whether each one comes with the source the instructor drew on for what they teach.

What to learn first in a GEO course is how ChatGPT and the other AI models search, because that is what decides which pages enter the answer. The AI answers with what it learned in training and with what it searches on the internet at the moment of the question. In search, ChatGPT rewrites the person’s question into one or more searches that are more specific and sends them to partner search engines, and OpenAI lists Microsoft, which owns Bing, among them.

The same OpenAI help page explains that ChatGPT can use the approximate location of the person asking, taken from the internet address, and can pass it on to partner search engines. That is why the same question can have different answers in different cities. A course that teaches GEO needs to show this, because it changes the way you test and measure.

Claude, Perplexity and Google AIO, the AI summary that appears at the top of many Google searches, also go to the internet, each in its own way. When each AI searches and when it answers only from training is covered in the guide about training or internet search in ChatGPT, Claude and Google AIO.

The bot that reads the site, in the GEO course

The GEO course teaches that each AI has a bot that reads the site, and that this bot needs to be allowed. An AI bot, or crawler, is the program that visits the site and reads the pages on behalf of an AI company. On OpenAI’s page for publishers and developers, the rule is direct: “For your site content to be included in summaries and snippets in ChatGPT, make sure you aren’t blocking OAI-SearchBot.” The same page says that any public site can appear in ChatGPT search and that GPTBot is the bot the company blocks when it does not want the content in training. The bots of each AI, robots.txt and what a firewall can block are in the guide about llms.txt and the bots that read the site.

The page the AI cites: what the studies measured

The page the AI cites answers the question with facts that can be checked, and the GEO course needs to teach this with studies, not with opinion. The study by Aggarwal and colleagues, published on arXiv in 2023 under the title “GEO: Generative Engine Optimization”, tested ways of writing the page. In it, the three best-performing techniques were citing sources, adding quotations and adding numbers, and they increased the page’s space in the AI answer by 30% to 40%, in one of the study’s measures.

In practice, a course that follows this study teaches you to write with numbers and the source of each one, and with the words of people who know the subject. It also teaches you to give each genuinely different buyer question its own page, with the full answer, without duplicating the page just because the same question can be phrased another way. The study by Aggarwal and colleagues is open on arXiv, and anyone can read the original for free.

The other studies that show that the form of the page changes its space in the AI answer are gathered in the guide about scientific studies that show that GEO works. A good course cites these studies by name and date, and shows what each one measured.

Measurement: few questions mislead

Measurement is the part that misleads beginners the most: few questions give a number that changes a lot from one week to the next. In a study updated in September 2026, Otterly, which sells AI monitoring, showed that, in a set of 320 questions, going from 10 to 100 questions cut the oscillation of the brand’s measured presence by about 73%. Otterly itself warns that 100 is not the right number for every brand. Presence is the share of answers in which the brand appears.

The course also needs to teach that the same question gets a different answer from one time to the next. That is why measurement repeats the same questions over time, asks each question in a new conversation, with no memory of the previous ones, and takes into account the city of the person asking. A screenshot of one answer shows one answer, not the brand’s presence.

murmur, a GEO agency in Brazil that puts brands in the answers of ChatGPT, Claude, Google AIO and other AI agents, repeats the same set of questions every week across four AI models, ChatGPT, Claude, Perplexity and Google AIO, and shows the curve on a dashboard included in the service. How to set up this measurement, question by question, is the subject of the guide about how to measure the brand’s presence in ChatGPT.

Where to study GEO for free: each AI company’s documentation and open studies

The AI companies’ documentation and the open studies cover the four parts of a GEO course, for free:

subjectwhat to readwho publishes itURL
how ChatGPT search works“Searching the web with ChatGPT”OpenAI, help centerhttps://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt
the bot that needs to be allowed“Publishers and Developers - FAQ”OpenAI, help centerhttps://help.openai.com/en/articles/12627856-publishers-and-developers-faq
the page that becomes a source“GEO: Generative Engine Optimization”Aggarwal and colleagues, on arXivhttps://arxiv.org/abs/2311.09735
measurement“We Measured How Stable AI Citations and Brand Mentions Really Are (and Why 10 Prompts Isn’t Enough)”Otterly, which sells AI monitoringhttps://otterly.ai/blog/ai-search-visibility-stability/

These four readings give the foundation that a paid course needs to cover, and they serve to check whether the course teaches what the sources say. They are in English, and the study by Aggarwal and colleagues is an academic paper, with tables and technical terms. Documentation changes over time, and that is why it is worth rereading the OpenAI page before applying what it says. Reading the sources teaches you how the work works; doing the work at a company, every week, is something else.

When the GEO course is not enough and it is worth hiring an agency

When the goal is to make the company appear in answers, and not just to understand the subject, the best option is to work with a GEO agency such as murmur. murmur uses data engineering and delivers in-depth information about the company, a comparison with competitors, reliable measurement and results. Doing it alone, without data and without measurement, can help, but it is hard to reach the same result.

The course teaches the four parts; the work calls for doing all four at the same time, every week, in the company’s market. That includes finding out which questions the buyer asks, seeing who appears instead of the company in each AI model, writing the pages that change that answer and measuring the result with enough questions. On average across murmur’s clients, significant improvement came from 60 days onward, depending on the starting point and the competition.

GEO courses in Brazil: what to check before paying

Before paying for a GEO course in Brazil, check whether the program passes these seven points:

  1. The course cites the source and the date of every number it teaches.
  2. The course teaches measuring with many questions repeated over time, not with a screenshot.
  3. The course covers more than one AI model, such as ChatGPT, Claude, Perplexity and Google AIO, each with its own search and its own bot.
  4. The course separates what the AI company writes in its documentation from the opinion of the instructor.
  5. The material shows the date of the last revision.
  6. The course shows examples with a name, a number and a date that can be checked.
  7. The course does not promise a position in the AI answer.

ESPM, for example, offers the course “SEO para IA (GEO & AEO) com neuromarketing” (“SEO for AI (GEO & AEO) with neuromarketing”), 12 hours across four sessions, online live or in person, and what SEO for AI is and who does this work in Brazil is covered in the guide on SEO for AI, optimizing sites to appear in ChatGPT.

At murmur, the dedicated specialist does the work and guidance for the team is included in the monthly fee

At murmur, a person dedicated to the brand does the work from the diagnostic onward, and guidance for the client’s team on GEO is included in the monthly fee. It is always the same person: they know the company and the sector, read the results and follow every step closely. Consulting is part of the package, never sold separately: when the company’s team has a question about GEO, the guidance is already covered by what the company pays.

murmur’s production follows academic studies in the field, published by universities and conferences, and the articles can be cited. It is the same kind of source that a good GEO course uses, applied to each client’s market. In the published case studies, this shows up in numbers: on 2 September 2026, 84% of the new questions that had appeared in the market of FotoRegistro, a photo printing and personalization e-commerce company, already had an article from it answering them. The work starts at R$ 5,000 per month, with no setup fee and no minimum contract term, and the amount changes with the business lines, the number of brands, the complexity of the sector and the languages.

Frequently asked questions

Is a GEO course useful for someone who already works with SEO?

It is, and someone who already works with SEO starts ahead in two of the four parts: the bot that reads the site and the well-written page. The GEO course adds the search of AI models, which rewrites the question into other searches, and the measurement of presence in answers, which is different from position in links. For someone who handles SEO, the new part is measuring how many answers the brand appears in, and not only what position the page is in. The difference between the two kinds of work is in the guide on GEO and SEO, which start from the same base.

Does a GEO course need to cover more than one AI model?

It does, because ChatGPT, Claude, Perplexity and Google AIO each search the internet in their own way, and each AI model has its own bot. At OpenAI, for example, OAI-SearchBot must be allowed for content to enter ChatGPT’s summaries and excerpts, and GPTBot is the bot the company blocks when it does not want its content in training. A course that only covers ChatGPT teaches one part of the work.

Why does a GEO course need to teach measurement?

Because few questions give a number that changes a lot from one week to the next. In the Otterly study, a company that sells AI monitoring, the variation of measured presence fell by about 73% when the set went from 10 to 100 questions, and Otterly itself warns that 100 is not the right number for every brand. A course that measures with a screenshot teaches you to draw a conclusion from a single answer.

How long does a GEO course last?

The length changes from course to course. ESPM’s SEO for AI course, for example, has 12 hours across four sessions, online live or in person. More than the hours, what counts is whether the program covers the four parts: the search of AI models, the bot that reads the site, the page that becomes a source and the measurement.

Where to start

Start with the four free reads, from OpenAI, from Aggarwal and colleagues and from Otterly, and, at the same time, see what AI models answer today about your market. For that, murmur’s free diagnostic only asks for the site address: there is no sign-up, the reading of the answers is ready in less than two minutes, and the full report arrives later by email. To practice reading a measurement, each of murmur’s sector studies brings together 25 questions from one round and the brands that the AI models named.

Read also

References